Group-wise optimization and individualized prediction of structural connectomes

Construction and modeling of structural and functional connectomes from neuroimaging data have shown great promise in elucidating the fundamental architectures of the human brain. In this paper, we present a novel framework to optimize large-scale cortical landmarks by maximizing the structural connectome agreement across a group of subjects and then use group-wise consistent connectome as an effective constraint to predict these optimized landmarks on new individuals. This cortical landmark optimization and prediction framework have been developed, validated and applied to the publicly available Dense Individualized and Common Connectivity-based Cortical Landmarks (DICCCOL) system as a test-bed with large testing samples (N=120). The experimental results suggest that our framework can substantially increase the group-wise connection consistency between DICCCOL landmarks across individuals' brains. After applying our framework, the anatomical and connectional profiles of those landmarks are remarkably improved, thus offering a solid structural foundation for future investigation of a variety of brain sciences questions.